[Mlir-commits] [mlir] [mlir][xegpu] Convert Vector contraction to XeGPU (PR #122115)

[Chao Chen]

https://github.com/chencha3 commented:

I think the code itself looks good to me. A concern I have (actually you have pointed out in the PR) is the size. XeGPU code is HW specific. From design perspective, I believe we need (and will) to enhance it with some arch parameter to verify whether the code is valid for that arch. For downstream, we currently has an arch-aware blocking pass for xetile before lowering xetile to xegpu, and the lowering pass has an arch-specific target to verify the generated xegpu code is valid for that arch. Sounds like you will have the same thing for vector.contract before lowering it into XeGPU? (unfortunately, we don't have this target for upstream yet) 

https://github.com/llvm/llvm-project/pull/122115